Initially, John Knoll had naïve ideas of what creating the Maelstrom would require. All Pirates images © Disney Enterprises Inc. All rights reserved. 

Once again, VFXWorld asks the nominated supervisors what they thought of each other’s work as well as their own. Industrial Light & Magic’s Scott Farrar and John Knoll (last year’s winner) pushed the boundaries of hard surface rendering and fluid simulation with Transformers and Pirates of the Caribbean: At World’s End, respectively, and Mike Fink oversaw one of the season’s biggest CG undertakings with The Golden Compass.

Bill Desowitz: In a year dominated by so many fluid simulation achievements, talk about the significance of the Maelstrom?

John Knoll: I knew we were going to be dealing with heavy amounts of CG water, and partly because I previously tried to minimize it, I had little experience with CG water. And going into the Maelstrom I had naïve ideas of what it would require. My perception was that, given all the development work that went into Poseidon, which looked great, I should be able to take those tools and just use them. But as we really started getting serious about making an image, I discovered that there were all sorts of limitations with the tools and that there were things we needed to do that were unique to Pirates.

BD: Such as?

JK: I’ll give you a couple of examples. The fluid simulations that we had done on Poseidon were of a flat ocean. And so imagine, for the most part, that the simulations fit into a volume that is very large in two dimensions and very shallow in the third dimension. But imagine the shape of the Maelstrom being very large in all three dimensions. And so this work is done with 3D voxels: a giant vector field that takes a significant amount of memory to represent a shape like that. And so we were going to have trouble getting the kind of detail that we needed in our fluid simulation, even given the 32-gig machines that we had to do this. So a lot of effort went into maximizing the amount of detail and optimizing our use of memory and computer resources. Another thing that hadn’t occurred to me before we went into this was the secondary level of detail that’s added to water that is procedural -- deep water that gives you small ripples and fine scale waves. And a lot of times addressing that into a shot is based on what the camera is actually seeing. And so there are resolution-dependent optimizations that are made because of the curved shape of the Maelstrom, that you’re seeing many different scales of water simultaneously. And some of those few dependent optimizations were not possible. We also had to direct the water to do what the script was demanding and not necessarily what the fluid simulator wanted it to do. And I’ll give you an example of that. An important story point to Gore [Verbinski] was that the Maelstrom spins faster when you go deeper down into it. And they use this property in the script a couple of times. The Flying Dutchman cuts down deeper into the Maelstrom where it’s spinning faster to catch up to the Black Pearl and ends up on the Black Pearl’s tail, and the Black Pearl has to dive down deeper into the Maelstrom to [elude] the Flying Dutchman. So it was important to Gore that we see that. So even though the Maelstrom was taking on the correct shape of what’s called an irrotational vortex, the kind that’s formed by water going down a drain, if you frame up a shot in 2:35.1 frame, the water wasn’t visibly fast enough down the bottom of that frame. So we wrote tools to do deformations on top of fluid simulations. So we had a twist deformation in that case where we could spin the lower part of the frame so that it was going noticeably faster to make that story point. Then because we wanted to do hundreds of shots and maximize the reusability of these expensive simulations, we decided that the wakes of the boats should be simulated separately from the main fluid simulation of the Maelstrom so that we could put the boats wherever they needed to go for a particular shot, and then add the weight around the boat in that position. And so here you have two fluid sims that are these giant 3D vector fields and we needed to have a way of doing compositing with vector fields, so we wrote some tools that allowed us to do these fluid sim composits.